https://osf.io/fk3xr/

tl;dr

Interesting to note the study has been withheld from publication while the APA figures out if it's ethical or not, despite the fact that Stanford U already said it was using APA guidelines. Apparently it's caused a bit of a kerfuffle among certain groups. Irony is 'lol open science' anyone can see it in its final form anyways.

Just from glancing thru it, saw this bit here. My guess from this is that the AI works by parsing out masculine and feminine facial features, and when a man is too feminine he's classified as gay and v/v.

The results show that the faces of gay men were more feminine and the faces of lesbians were more masculine
than those of their respective heterosexual counterparts.